A METHOD OF TRUST MANAGEMENT IN WIRELESS SENSOR NETWORKS

International Journal of Security, Privacy and Trust Management (IJSPTM) Vol 7, No 3/4, November 2018
DOI : 10.5121/ijsptm.2018.7401 1
A METHOD OF TRUST MANAGEMENT IN WIRELESS
SENSOR NETWORKS
Janusz Górski and Alan Turower
Department of Software Engineering, Gdańsk University of Technology, Gdańsk, Poland
ABSTRACT
The research problem considered in this paper is how to protect wireless sensor networks (WSN) against
cyber-threats by applying trust management and how to strengthen network resilience to attacks targeting
the trust management mechanism itself. A new method, called WSN Cooperative Trust Management
Method (WCT2M), of distributed trust management in multi-layer wireless sensor networks is proposed
and its performance is evaluated. The method is specified by giving its class model in UML and by
explaining the related attributes and methods. Different attacks against the network and against WCT2M
deployed in the network are considered. The experimental evaluation of WCT2M involves laboratory
experiments and simulations using a dedicated simulator. The evaluation focuses on efficiency of detecting
and isolating the malicious nodes that implement different attack scenarios in the network and on the
method’s sensitivity to the changes in effectiveness of the security mechanisms deployed in the network
nodes.
KEYWORDS
Wireless Sensor Network; trust management; security threat; attack scenario; simulation
1. INTRODUCTION
Wireless sensor networks (WSN) are important in different domains, including healthcare,
defense, security, environment monitoring and others. Although WSNs share a number of
security requirements with traditional networks, due to their specific limitations they may need
different solutions. WSNs have limited resources such as memory, power supply and
computational ability. Resource limitations stem from the need of tiny and low cost devices. This
restricts applicability of conventional protection mechanisms (like advanced crypto schemes) and
excludes complex and resource-consuming security solutions [1]. Moreover, the protocols used in
WSNs are often designed without devoting much attention to security requirements. ZigBee [2],
one of the most popular specifications for WSN communication protocols, may serve as an
example. ZigBee is built upon the IEEE 802.15.4 standard [3] which defines the Physical and
Media Access Control (MAC) layers for low cost, low rate personal area networks.
Figure 1 presents an example of a multitier WSN - this case study was borrowed from the
ANGEL project [4]. The base station is labelled BS and the nodes are labelled by natural
numbers. Solid lines represent the routes leading from nodes to the base station. In addition, the
dotted lines connect nodes being in their direct range. For example, node 2 is able to
communicate directly with nodes 1 and 3, and node 1 is its router to communicate with BS. The
nodes of this network are divided into the following clusters: cluster_1 = {1, 2, 3, 4}, cluster_2 =
{5, 6, 7}, cluster_3 = {8, 9, 10} and cluster_4 = {1, 8, 5}. Cluster_4 groups the heads of clusters
_1, _2 and _3 and together with the base station forms tier_1 = {BS, 1, 5, 8}. The remaining
nodes form tier_2 = {2, 3, 4, 6, 7, 9, 10}. The nodes of tier_1 (except the base station) are routers
– they forward messages from BS to their child nodes and from their child nodes to BS. The level
of a node is the shortest distance (in hops) from this node to the base station. The levels of the
nodes are as follows: for nodes 1, 5 and 8, level = 1 and for nodes 2, 3, 4, 6, 7, 9 and 10, level = 2.

2
The height h of a network is the maximal level of nodes of the network. The height of the network
of Figure 1 is h = 2.
Figure 1. An example two-tier network
Let us assume that the network applies fully synchronized sleep pattern [5] as presented in Figure
2a [6]. Node operation is split into sleep (labelled: T) and wakeup (labelled: A) periods. The
arrows represent the fastest scenario of sending a message from a node of level 2 to the base
station.
a) Fully Synchronized sleep pattern b) Spam attack
c) Black hole attack d) Message modification attack
Figure 2. Fully Synchronized sleep pattern and related attacks
The network as presented in Figure 1 can be subjected to various attacks. For instance, a
malicious router node, say node 8, may modify the forwarded messages or even drop them
selectively. In more complex situations the malicious nodes can involve themselves in even more
advanced attack schemes.
An alternative that can help to cope with the above problems is trust management. Trust can be
defined as a bet that those entities, which you cannot control, will act in a predictable manner that
is favourable to your cause [7]. Generally, trust is a relation between a trustor (the trusting
subject) and a trustee (the subject being trusted). The objective of trust management in WSN is to
distinguish between trustworthy network nodes and untrustworthy ones. Then, the trustworthy

3
nodes can cooperate to provide trustworthy network services and the untrustworthy nodes are
excluded from the network [8].
Trust management schemes are proposed as a solution to many security issues in WSN. For
instance, [9] asserts that cryptography and strong authentication schemes are not a solution to
WSN security issues because they do not detect a large set of attacks such as selfish behaviours
and black holes while at the same time their implementation at low cost is not feasible. They
survey trust models in an attempt to explore the interplay among the implementation
requirements, the resource consumption and the achieved security.[10] emphasizes the
importance of trust as an important issue of security schemes in wireless sensor networks. They
categorize various types of attacks and countermeasures related to trust schemes in WSNs. They
also present an extensive literature survey by summarizing state-of-the-art trust mechanisms in
two categories: secure routing and secure data. [11] lists the best practices being essential for
developing a good trust management system for WSN. They find that the nature of WSN and its
vulnerabilities to attacks makes the security tools required for them to be considered in a special
way. [12] proposes a special Trust and Reputation layer added to the modified version of the OSI
model, under Ad hoc On-demand Distance Vector (AODV) layer. This approach allows
providing intrusion tolerant routing in WSNs. The Trust and Reputation layer uses a proactive
approach, where a potentially not trusted node is explicitly tested to obtain a direct measurement
of corresponding trustworthiness. This allows collecting relevant information by the trust-based
architecture independently of the specific features of the considered routing protocol. [13] pays
attention to the problem of collusion attacks in trust-based MANET networks (mobile ad hoc
networks). Trust management may be faked by cheaters: several nodes collude in order to rate
each other with the maximum value and at the same time decrease other nodes’ reputation by
giving negative recommendations about the latter. The authors propose detecting the colluding
nodes by calculating recommendation deviations and punishing these nodes by discarding them
from further communication.
To strengthen protection against various attack schemes we propose a new trust management
method called WCT2M which is implemented by a software package to be installed on each node
of the network. WCT2M assumes that each node N has a potential to detect that an incoming
message sent by another node, say M, violates the security policies implemented by N and this
information becomes available to WCT2M together with the M’s identity (no assumptions about
the ‘strength’ of such security policies are being made). The information that a malicious message
has been received from M is used by N to compute the trust level assigned to M. In addition, the
trust level assigned to M by node N takes into account the trust levels assigned to M by other
nodes and communicated to N as recommendations. Then, node N communicates to other nodes
(in a form of recommendation) the computed trust level assigned to M and depending on this trust
level N decides whether it continues its cooperation with M or not. Together with the trust levels
assigned to its neighbouring nodes, N maintains historic data related to the previous assessments
of the messages received from M (called validity history) and the deviations between the own
assessment of trustworthiness of M and the recommendations related to the trustworthiness of M
received from other nodes (called deviation history). Both, the validity history and the deviation
history are taken into account by N while computing a new trust level of M, before the
recommendation related to M is being sent to other nodes. The functionality of WTC2M software
package was specified using UML and then this specification was used to implement the package.
This implementation was used to demonstrate WTC2M in a laboratory environment and to run
experiments of a limited scale. To provide for analysing performance of the method in larger
networks that are subjected to various attacks (against the network and against the trust
management mechanism itself) we developed a dedicated simulator. During simulations we
assumed a probabilistic model for both, the effectiveness of mechanisms enforcing the security
policies implemented by the network nodes and the attacks against the analysed network and its
trust management mechanisms.

4
The remainder of this paper is organized as follows. In section 2 we review the related works and
in section 3 we introduce a range of different attacks against a network and the attacks against a
trust management mechanism deployed in the network, and state the problem we are dealing with
in this paper. In section 4 we introduce WCT2M, a new trust management method for WSN and
describe the way of method instantiation and the parameters and data types used by the method.
Section 5 gives more details of WCT2M and in particular presents the class model and the
execution model of the method. In section 6 we present the assumed evaluation strategy, the tools
used during evaluation and the scope of evaluation experiments. In section 7 we present the
results of the evaluation experiments and section 8 present the discussion of the results achieved
and the directions of further research.
2. RELATED WORKS
Trust management for complex wireless sensor networks is currently an area of active research.
The main question is how network nodes should behave in order to effectively and efficiently
identify and isolate malicious nodes in the network. There are two main models considered for
trust management: centralized and distributed [1]. The centralized model distinguishes the Trust
Authority (this could be a role of the base station) which manages trust relationships between
nodes [14]. This solution is efficient and manageable, but it has problems with scalability and
robustness. The distributed trust model is considered to be suitable for large-scale sensor
networks. [1] finds this model appropriate for sensor network security design because a node
focuses on the trustworthiness of its neighbors and assesses if these nodes obey the agreed
security policies. They propose a corresponding security framework with different security
schemes. However, their work does not take into consideration limited resources of nodes in
sensor networks. [15] proposes a distributed agent-based trust management scheme where each
agent node independently monitors the behaviour of the nodes within its radio range and
broadcasts their trust ratings. They also introduce a reputation based trust model using
probability, statistics and mathematical analysis and have suggested a trust system to build up a
reputation space and trust space in WSNs [16].
A hybrid approach is also possible. It combines the advantages of the centralized and distributed
models (but also can incorporate problems connected with each of these architectures). The
network structure comprises two-levels where all nodes are divided into clusters and each cluster
head is an element of so called backbone network, which enables communication of cluster heads
with the base station. [17] proposes a trust and reputation management scheme that uses mobile
agents running on each node. In this model there is a central agent launcher responsible for
generating and launching agents into the network. However, there is no central repository of trust,
which makes trust exchange significantly more difficult. [18] explore the Dempster-Shafer
Theory of Evidence (DST) as part of a framework called DST-SDF and discuss some of its
advantages and disadvantages. It differs from existing reputation schemes in that the well-known
but in their opinion faulty watchdog mechanism is dispensed with, and end-to-end
acknowledgments are used instead. [19] states that trust based only on reputation is not sensitive
enough to perceive suddenly spoiled nodes that may launch intelligent attacks against a trust-
establishment mechanism. It presents a Trust Model based on Risk evaluation (TMR) to
effectively deal with conflicting behaviours of malicious nodes. However, risk factor evaluation
causes the trust model less reliable because it becomes more sensitive to on-off attack. [20] states
that encryption and authentication can prevent the external attacks on Wireless Sensor Networks,
and trust management schemes are better approaches for defending against the internal attacks in
WSN. They describe the internal reputation time-varying attack, in which the reputation value of
nodes is manipulated to adjust dynamically by the compromised nodes or a malicious attacker.
The authors propose a Time-window-based Resilient Trust Management Scheme (TRTMS),
based on BETA distribution. In the scheme, the behaviours of compromised nodes are analyzed

5
for a period, and then the difference judgment and the trend analysis are utilized to identify the
abnormality of nodes’ reputation.
A distinguishing feature of WCT2M is that it is implemented on top of the stack of protocols and
remains neutral with respect to the protocols used in the network. The method can coexist with
any security mechanism that implements specific security policies at network nodes to assess
validity of the incoming messages. WCT2M makes no initial assumptions concerning the
effectiveness of such assessment. It simply uses the result of validity assessment as a factor
influencing the trust value (also called reputation) assigned to the sender of the incoming
message. The other factors influencing this trust value are the recommendations received from the
neighbouring nodes and the validity and deviation histories related to the sender of the message.
3. PROBLEM STATEMENT
WSN networks are subjected to various attacks and, if protected by a trust management
mechanism, they can also be exposed to the attacks targeting the trust management mechanism
itself.
3.1. NETWORK ATTACKS
Three different attack models were chosen to demonstrate possible threats to WSN. They are
representative for a number of more concrete network attack scenarios. We describe them below
referring to the example network of Figure 1.
3.1.1 Spam attack
Spam attack happens when unnecessary and useless messages are generated and spread over the
network. This attack is intended to waste network resources (energy, bandwidth), especially of the
routers. This type of attack can quickly isolate router nodes from the rest of the network. In this
way the whole network may be destroyed, despite the fact that most nodes remain operational.
Attack scenario: The attacker node, in its wakeup period, sends multiple messages to the base
station. Figure 2b illustrates the spam attack – in its wakeup period, a node on level 1 of the
network sends more messages than it was supposed to do. Attack success criteria: The attack is
considered successful if the attacker continues to deliver spam messages and remains undetected.
3.1.2 Black hole attack
In its simplest form the black hole attack involves losing the received messages to prevent their
further propagation. Such situation is not unusual even during a normal operation of sensor
networks – a network must be ready for sudden disappearance of a node, for instance due to the
exhaustion of its energy source. Therefore, most protocols have a mechanism that periodically
checks all paths, and if a path stops working, it becomes re-created. This action eliminates
malicious nodes from the path (if any) and prevents them from sabotaging the network operation.
For this reason, to avoid detection, black hole attacks use a mechanism of random transmission of
packets to confuse the maintenance mechanism. Another variant can be the selective forwarding
attack where a malicious node selectively drops sensitive messages [21]. These attacks target
routers because only router nodes forward messages and can drop them becoming a black hole.
Attack scenario: The attacker node does not forward any messages or forwards only some of them
(to the base station and to other nodes). All not-forwarded messages are dropped. Figure 2c shows
the black hole attack led by a node on level 1. A message which has been dropped in the black
hole is represented by a dotted arrow. Attack success criteria: The attack is considered successful,
if a malicious node located on at least one routing path blocks forwarding of messages and
remains undetected.

6
3.1.3 Message modification attack
Message modification attack targets router nodes and involves unauthorized modification of a
message data and then resending the message. The attacker can modify the content, the
information about the receiver node and/or information about the sender node. Attack scenario:
The attacker node modifies all or some messages being forwarded to the base station or to other
nodes. All other messages are forwarded without any modification that contradicts the network
policy (some message attributes can change during ‘normal’ forwarding). Figure 2d illustrates a
message modification attack – the message that was maliciously modified by a node on level 1 is
represented by a solid arrow. Attack success criteria: The attack is considered successful if a
malicious node located on at least one routing path modifies messages and remains undetected.
3.2. TRUST MANAGEMENT MECHANISM ATTACKS
Trust management helps to make the whole network more resistant to attacks, but it can be the
target of attacks itself. Two popular attacks on trust management mechanism are described below
[22].
3.2.1 Decreased frequency attack
To confuse the trust management mechanism an attacker can decrease frequency of its attack
against the network. If malicious actions are performed rarely enough, the malicious node can
regain its trustworthiness before being detected and still remains in the network.
3.2.2 Collusion attack
Trust management mechanism can get confused by collusion of malicious nodes. Some nodes can
collude in order to mutually increase their trustworthiness and at the same time decrease the
trustworthiness of other (innocent) nodes [13].
3.3. DEPENDENCE ON EFFECTIVENESS OF SECURITY MECHANISMS
Security depends on trust (for instance in public key infrastructures) however, trust can also be
built on top of security. Consider a WSN node with some security mechanisms on board which
implement a chosen security policy. If an incoming message violates this policy, it consequently
decreases the trust in its source (the message originator) and this information is then spread to
other network nodes in a form of recommendation. Such recommendation can affect trust of these
other nodes in the message originator even if they were not the direct recipients of the spoiled
message. The resulting changes of trust eventually lead to the exclusion of the sender of the
spoiled messages from the network. A question arises to what extent the efficiency of trust
management (understood as a time delay needed to exclude the malicious nodes from the
network) depends on the effectiveness of the security mechanisms installed on network nodes. Of
particular interest is the situation where we still can have efficient trust management even if the
effectiveness of the security mechanisms is relatively low (note that less effective security is
usually cheaper in terms of resource consumption, which could be an attractive alternative for
WSNs).
3.4. THE PROBLEM
The research problem considered in this paper is how to protect a WSN equipped with incoming
message validity checking mechanisms of limited effectiveness deployed on the nodes, against
various attack scenarios, by applying a trust management mechanism that is resilient to typical
attacks targeting this mechanism.
To solve the above problem we proposed a new method of trust management in multi-layer
wireless sensor networks called WCT2M (WSN Cooperative Trust Management Method). We

7
focused on clustered networks to limit performance problems related to network expansion, as
clusters allow achieving high scalability and high energy efficiency of large wireless sensor
networks [23].
The main contribution of this paper is WCT2M, a new method of trust management in multi-layer
clustered wireless sensor networks. The method implements a distributed trust management
model. A laboratory implementation and a dedicated simulator have been developed and used to
evaluate WCT2M efficiency in protecting against selected attacks in networks of realistic size.
4. WTC2M – A METHOD OF TRUST MANAGEMENT IN WSN
WCT2M implements a distributed trust management model. No central entity is being assumed
after a WTC2M enabled network starts its operation. The trust related decisions made at a node
are solely based on local assessment of the received messages and on the recommendations
received from other nodes.
4.1. INSTANTIATION OF WCT2M
WCT2M is implemented by a software package called WCT2M software. To run WCT2M
software, the network administrator needs to perform the following steps:
Step 1: Install WCT2M software on each network node, including the base station, and assign
values to the parameters of the method;
Step 2: For each node, configure access of WCT2M software to the routing information (to
provide for distinguishing the base station) and to the synchronization protocol run by the
network;
Step 3: For each node, interface WCT2M software to the communication software installed on the
node to provide access to the incoming messages and to the results of their assessment by security
mechanisms enabled on the node;
Step 4: Start the network and allow propagating trust information between its nodes;
To join a new node to the existing network, the network administrator needs to follow steps 1-3
with respect to the node and then the new node can join the running network.
4.2. PARAMETERS OF WCT2M
In a WCT2M enabled network each node acts in a dual role: as a trustor (for incoming messages
it makes real-time decisions if a sender can be trusted) and a trustee (for outgoing messages other
nodes judge if the sender can be trusted).
To represent trust, all nodes running WCT2M software use the same scale called trust scale.
There are three characteristic values related to this scale: full trust – means that the node is fully
trustworthy, initial trust – is the initial credit given to a node and cut-off – is the trust level below
which the node is considered untrustworthy.
Each network node participates in the trust management process and maintains data on its trust in
other nodes. The corresponding data structure maintained at each node is called trust table.
Each node can communicate its trust in other nodes by sending recommendations.
Recommendation is an entry of the local trust table sent to other nodes. A node can recommend
any other node except itself.
While receiving a message, the node makes decisions based on trustworthiness assessment of the
sender of the message and the message itself. The assessment is based on two complementary
approaches: Policy-based approach – the trustor evaluates trustworthiness of the trustee assessing

8
its observed behaviour and its conformance with agreed policies, notably the security policy, and
Reputation-based approach – the trustor takes into account recommendations from other nodes if
they trust the trustee. The influence of these two viewpoints is controlled by the Cooperation
Factor (CF), same for all nodes in the network. The value of CF is being defined during Step 1 of
WCT2M instantiation. CF assumes values from 0 to 1, where 0 means that recommendations are
discarded in calculations and 1 means that the trust assessment is solely based on
recommendations.
Trust value assigned to the sender of a message and maintained by the receiver of the message
can change in effect of the assessment of an incoming message. If node B sends a message to
node A and the message violates the security policies assumed by A, B’s trust value maintained
by A is decreased by changenegative factor:
,
and in case the message agrees with the security policies at the receiving node, the trust value of
the sender increases by changepositive factor:
where both, changenegative and changepositive are real numbers belonging to [0, 1]. WCT2M software
running on a node can call a service returning the result of security assessment of an incoming
message. This service is not a part of WCT2M software. WCT2M assumes that the result of this
assessment is a binary value where 0 means that the message violates the corresponding security
policy and 1 means the opposite. At each node, validity history is a data structure representing, for
each sender node, the history of such evaluations of the incoming messages. The validity history
is implemented as a FIFO queue of fixed size, called validity history table.
In addition to its own assessment of the trustworthiness of its neighboring nodes, each node
receives recommendations representing trust assessments from the other nodes’ perspective.
WCT2M assumes that each node compares such recommendations with the own assessments and
the results of such comparisons are represented in a data structure called deviation history. At
each node and for each neighbouring node, the deviation history is a queue of (binary) values
indicating if the recommendation received from the neighbouring node deviates from the own
trust assessment of the recommended node. The deviation history is represented at each node as a
FIFO queue of fixed size, called deviation history table.
5. METHOD DESCRIPTION
This section gives details of WCT2M software structure and operation.
5.1. CLASS MODEL OF WCT2M
The class model of WCT2M software is presented in Figure 3.

9
Figure 3. WCT2M class model
The model consists of the following classes (the entities stored at the node executing a given
method of the specified class are prefixed by ‘own’, for instance ‘own trust table’ refers to the
trust table stored on the node that executes the method):
• Class Message - represents a message (data attribute) exchanged between nodes (source
and receiver attributes)
• Class TrustTable - represents trust values assigned to other nodes by a given node. It
allows to execute the following methods:
decreaseTrust(Node) - assigns a new, decreased trust value (calculated from the
current trust value) to Node in the own trust table;
getTrust(Node) returns current trust value assigned to Node in the own trust table;
increaseTrust(Node) assigns a new, increased trust value (calculated from the current
trust value) to Node in the own trust table;
recalculateTrust(List<TrustTable>) updates the values stored in the own trust
table based on the recommendations (the trust tables) received from the neighbouring nodes.

10
• Class TrustTableEntry - represents a single entry of TrustTable. It indicates a node
and a trust value assigned to it.
• Class Node - represents a node of the network (including the base station). It stores the
following attributes (the parameters assigned during Step 1 of instantiation of the WCT2M
are underlined):
cutOffLevel – cut-off level of the trust scale;
deviationCheck – auxiliary variable storing the deviation check result;
deviationRatingThresholdLevel1 – the primary (level 1) deviation rating
threshold used to control the number of values inserted to DeviationHistoryTable;
deviationRatingThresholdLevel2 – the secondary (level 2) deviation rating
threshold used to control the number of values inserted to DeviationHistoryTable;
messagesToSend – list of Messages to be sent;
neighbours – list of own neighbors;
trustDeviationThreshold – parameter used to calculate values inserted to
DeviationHistoryTable;
trustTables – auxiliary variable storing a list of recommendations (trust tables received
from other nodes);
trustValidityRatingThreshold – threshold used to control the number of values
inserted to ValidityHistoryTable;
Class Node allows to execute the following methods:
addNeigbour(Node) - adds a given Node to the own neighbours list.
isMessageValid(Message) - communicates with the security mechanisms
interfaced to WCT2M software during instantiation (configured during Step 3 of
instantiation of WCT2M) and checks if a given message is valid in the light of the
security policies implemented at the own node;
isSenderTrusted(Node) - checks in the own trust table if Node is considered
trusted (its trust value is above the cut-off level);
prepareRecommendations() - prepares a new Message containing own trust
table to be sent as a recommendation and adds it to the messagesToSend list;
processMessage(Message) - processes the incoming Message in accordance
with own objectives;
queueMessages() – adds messages that the node needs to send (the new messages
created by the node and/or the messages received by the router node to be forwarded)
to the messagesToSend list;
sendMessage(Node, Message) – sends Message to Node and removes it
from the messagesToSend list;
sendMessage(Message) - sends Message to all own neighbors and removes it
from the messagesToSend list;
waitForEndOfPhase() - pauses till the end of the current phase (execution of
this method depends on the synchronization protocol which is assumed for the
considered network).
• Class ValidityHistoryTable - each node maintains a separate copy of this table. For
each neighboring node, the table maintains the history of validity evaluations of the
messages received from this node. Results of such evaluations (positive or negative) are
stored in

11
a FIFO queue of a fixed size. If for a given node the number of negative evaluations exceeds
trustValidityRatingThreshold, the trust value of this node is decreased by calling
decreaseTrust(Node) method on the own trust table. Such check and the resulting
action (if needed) are being executed after every incoming message validity evaluation
performed by the receiving node.
The class allows executing the following methods:
getNumberOfNegativeEntries(Node) returns the number of False values in
the HistoryTableEntry of Node in the own ValidityHistoryTable;
update(boolean, Node) inserts to the own ValidityHistoryTable the
boolean value at the entry pointed to by Node (on the FIFO basis).
• Class DeviationHistoryTable - each node maintains a separate copy of this table. For
each neighboring node, the table maintains the history of deviations resulting from the
comparison of the own trust table with the trust table received as a recommendation from this
node. When node A receives a recommendation from node B, it calls
isDeviationAcceptable(TrustTableA, TrustTableB) where TrustTableA
is its own trust table and TrustTableB is the trust table received as a recommendation from
B. Then the deviation is calculated in accordance with the following formula:
where N is the cardinality of the intersection of two sets: the set of entries of A’s trust
table (the nodes known to A) and the set of entries of the trust table received from B (the
nodes known to B), represents the trust value assigned to node n in
TrustTableA and represents the trust value assigned to node n
in TrustTableB . If is below trustDeviationThreshold, the
deviation evaluation is True, otherwise is False. For each neighboring node,
DeviationHistoryTable stores the results of deviation evaluations in a FIFO
queue of a fixed size. If for a given node N, the number dev of False evaluations stored in
this queue satisfies the condition exceeds
deviationRatingThresholdLevel1 < dev deviationRatingThresholdLevel2
where deviationRatingThresholdLevel2 > deviationRatingThresholdLevel1,
the trust to the node is decreased by calling the decreaseTrust(N) on the own trust
table. If the number of negative assessments in the deviation history table exceeds
deviationRatingThresholdLevel2, the trust to the node is decreased twice by
two calls of decreaseTrust(N). This check and the resulting action (if needed) are
executed after receiving a valid (positively verified by
isMessageValid(Message)) recommendation from N.
The class allows to execute the following methods:
getNumberOfNegativeEntries(Node) returns the number of False values in
the HistoryTableEntry of Node in the own DeviationHistoryTable;

12
isDeviationAcceptable(TrustTable, TrustTable) calculates a
deviation between the received trust table and the own trust table and checks if it is
lower than trustDeviationThreshold;
update(boolean, Node) inserts to the DeviationHistoryTable of Node
the first parameter (boolean) on the FIFO basis.
• Class HistoryTableEntry - represents a single entry of ValidityHistoryTable
or of DeviationHistoryTable.
5.2. EXECUTION MODEL OF WCT2M
Execution of a WCT2M enabled network is structured into WCT2M cycles, each cycle divided
into two phases:
- Data phase: Each node sends/receives ‘regular’ messages to/from its neighbours and
forwards messages received from other nodes, if it is a router node. Based on the received
messages, the receiver node (the trustor) evaluates the sender nodes (the trustees) if they
can be trusted and then updates the corresponding trust values in its own trust table.
- Recommendation phase: Each node sends recommendations to its neighbours. The
recommendations are related to the end of the data phase (the trust table entries
exchanged as recommendations are the snapshots taken at the end of the data phase, after
all updates related to the incoming messages have been inserted). The receiver node (the
trustor) evaluates the sender nodes (the trustees) based on its own trust table and on the
incoming recommendation messages. If they are valid and can be trusted, the receiver
recalculates its trust table referring to the received valid recommendations (the
recommendations from untrusted nodes and the recommendations assessed as being
invalid are discarded).
The exact lengths of the data phase and the recommendation phase depend on the MAC protocol
used in the network, the sleep scheduling pattern adopted and the density of the network (two
interfering nodes cannot transmit at the same time because of message collisions [33], so
in dense networks recommendation exchange takes more time than in sparse ones). It
means that the time length of cycles can vary during the whole life of a network. The idea of
WCT2M cycles and phases is presented in Figure 4.
Figure 4. Cycles and phases of the WTC2M enabled network
WCT2M assumes that every node of the network participates in the synchronization protocol so
that it is able to synchronize with other nodes to work in such cycles. The method also assumes
that the synchronization protocol controls when a given phase ends and none of the actions
included in this phase lasts forever. If needed, a node can skip sending messages during a given
phase (to avoid delaying the phase termination).
6. EXPERIMENTAL EVALUATION
To assess WCT2M we applied the following approach: (1) Identified criteria and the
corresponding metrics for assessing the efficiency of WCT2M; (2) Developed a laboratory WSN
system to demonstrate feasibility of WCT2M (the laboratory system implements a small WSN of
Figure 1); (3) Because of scale limitations of the laboratory system, we used simulations to assess
the performance of WCT2M for larger networks. After reviewing the existing simulators

13
(including the assessment of their availability and cost), we decided to develop a dedicated
simulator. The laboratory system was used to calibrate the simulator and then the simulator was
used to evaluate WCT2M efficiency for different attack scenarios and different network sizes; (4)
A set of experiments has been carried out to assess WCT2M efficiency in detecting and isolating
malicious nodes implementing different attack scenarios and to assess how the effectiveness of
the security mechanisms deployed on network nodes influences WCT2M efficiency.
The laboratory network was created using elements of CC2520 Development Kit [24]. CC2520
is a Texas Instrument second generation ZigBee/IEEE 802.15.4 RF transceiver for the 2.4GHz
unlicensed ISM band.
The dedicated simulator, called WSN Cooperative Trust Management Simulator (WCTMS), was
written in NetBeans IDE [25], using Java 8 [26] with JGraph library [27]. WCTMS simulates the
execution model of WCTM where each cycle of the network (see Figure 4) corresponds to a
simulation turn during simulation run. Assuming that the cycles of the network have the same
time duration, the number of simulation turns can be interpreted as a measure of time delay in the
simulated network. And for a given simulation run and two different states of the simulated
network, S1 and S2, the number of simulation turns transferring the network from state S1 to state
S2 is called simulation distance between S1 and S2.
The simulator implements probabilistic model of network operation and of different threats
against the simulated network: pnd – is a probability of sending a message by a node in the data
phase of a cycle (nodes activity), pb – is the probability of sending a message by the base station
in the data phase of a cycle (base station activity), pnr – is the probability of sending the trust table
to the neighbors in a recommendation phase of a cycle, ps – is the probability of spoiling a
message by a malicious node, pe – is the probability of sending a spam message by a malicious
node, pd – is the probability of blocking a message by a malicious node, pch – is the probability of
modifying a message by a malicious node, pt – is the probability of modifying a trust table by a
malicious node.
In addition: r – is the probability of recognizing a spoiled message by the receiver node
(effectiveness of the security mechanisms installed on network nodes) and e – is the probability of
spoiling a message during transmission.
6.1. STRATEGY OF EVALUATION
A set of laboratory experiments was conducted using the network illustrated in Figure 1 and these
experiments were also repeated with the help of the WCTMS simulator to check if the results of
laboratory experiments and the simulations converge. Then an additional set of simulation
experiments was conducted using WCTMS. These experiments involved larger networks which
could not be directly implemented in the laboratory because of the limited resources. The
experiments were focused on evaluating the efficiency of isolating malicious nodes in the
network.
Metrics supporting the analyses were identified using the Goal-Question-Metrics (GQM)
methodology [28]. The overall goal of the experiments was defined as follows: analyse WCT2M
for the purpose of assessing the efficiency of malicious nodes isolation for different attack
scenarios. Below we give details of two metrics used in the experiments (for more details see
[29]): Average All Nodes Detected (AAND) [Real]: average number of WCT2M cycles needed to
detect all malicious nodes, calculated for all simulation runs of a given test case; Median All
Nodes Detected (MAND) [Real]: median number of WCT2M cycles needed to detect all
malicious nodes, calculated for all simulation runs of a given test case.
6.2. SCOPE OF EXPERIMENTS

14
The experiments were divided into groups. During simulations, the simulation distance was
interpreted as a measure of time delay between the initial state of the network and the state where
all malicious nodes have been successfully detected. The groups of experiments were as follows.
EXP1: Feasibility of implementing WCT2M: The objective of this experiment was to
demonstrate that WCT2M can be implemented in a typical laboratory environment. The
resistance of WCT2M to network attacks: spam attack, black hole attack, message modification
attack and decreased frequency attack was assessed experimentally. The laboratory experiments
were then repeated using WCTMS simulator to verify if the results of laboratory and simulation
experiments converge.
EXP2: Resistance to various network attacks (with possibly modified frequency of the attack):
The objective of this experiment was to examine the efficiency of malicious nodes detection and
isolation with respect to different network attacks assuming that the malicious nodes can modify
frequency of the attack to avoid detection. Each simulation run involved a network of up to 1000
randomly distributed nodes.
EXP3: Resistance to collusion attack: The objective of this experiment was to examine the
resistance of WCT2M to collusion attack. Each simulation run involved a network of up to 300
randomly distributed nodes.
EXP4: Influence of the effectiveness of security mechanisms: The objective of this experiment
was to examine how the effectiveness of security mechanisms implemented in the nodes impacts
the time delay of detecting the malicious nodes. Each simulation run involved a network of up to
300 randomly distributed nodes.
7. RESULTS OF EXPERIMENTS
In EXP1, the WCT2M software package was installed on the nodes of the laboratory network and
then numerous test cases were run, representing different attack scenarios. Then, each test case
was also repeated using WCTMS, with the same parameters and layout of nodes. For each test
case 100 simulation runs were executed and the results were compared with the results obtained
in the laboratory to verify if the laboratory and the simulation results converge. The experiments
demonstrated that WCT2M is feasible and can be implemented on real-world devices in a typical
WSN environment. The experiment results also demonstrated that WCT2M is able to protect
WSN against typical attacks and that the deviation history mechanism of WCT2M is an effective
tool to counteract the decreased frequency attack against the trust management mechanism.
In EXP2, the nodes were distributed in the square area (100 x 100 distance units) divided into 16
clusters of equal size (two-tier network). The base station was placed outside the nodes
distribution area. The node signal range was set to Zcl = 30 distance units for both, the base station
and the cluster heads, and Zn = 5 distance units for the remaining nodes. It was assumed that
during a given test case, the nodes do not change their positions.
The results for a network of 1000 nodes and for 10 malicious nodes are shown in Table 1.
Experiment EXP3 was to assess WTC2M efficiency while the validity history mechanism was
enabled. During experiments, the influence of the collusion attack was represented by parameter
pt (probability that the trust table gets modified in effect of collusion). The experiments were
conducted for networks of up to 300 nodes and for different numbers of colluding nodes (up to
the half of the network size). Each test case involved 100 simulation runs.
Table 1. EXP2 results (the deviation history mechanism on)
Test case Parameter value Metrics

15
MAND AAND
Spam attack pe = 80% 3.0 6.93
Black hole attack pd = 80% 15.0 29.70
Message modification attack pch = 80% 29.5 47.69
Figure 5 presents the result - the median time of detecting all colluding nodes. The underlying
assumption was that the probability of an adverse modification of the own trust table by a
malicious node pt=50%. The simulated network comprised 300 nodes.
Figure 5. Median delay of detecting all colluding nodes (validity history mechanism on)
The objective of EXP4 experiment was to assess efficiency of WCT2M depending on the
effectiveness of the security mechanisms implemented in the nodes (this effectiveness is
represented by r parameter in Table 2 and Table 3). For instance, r = 90% means that there is
90% probability that the security mechanism detects that an incoming message violates the
compulsory security policies. During this experiment we examined networks of different size (up
to 300 nodes) for different values of r (100%, 95%, 90%, 85%, 80%, 75%, 70%, 65%, 60%, 55%,
50%) and for different numbers of malicious nodes in the network. Figure 6 presents the MAND
metric for a network of 300 nodes (10 of which are being malicious).
Figure 6. Median time delay of detecting all faulty nodes
From Figure 6, it can be observed that the effectiveness of security mechanisms highly impacts
efficiency of WCT2M, if r drops below 65%.The experiment shows that if r 65%, the malicious
nodes are detected with the delay which is inversely proportional to the value of r. However, if
the value of r drops below 65%, the efficiency of malicious nodes detection decreases rapidly.

16
The explanation of this observation is as follows. The mechanism adopted in WCT2M for
changing the trust level in trust tables is not ‘linear’ what is presented in Figure 7. The trust
increase curve illustrates the increase of trust in effect of positive events (where a positive event is
this which affects trust positively, for instance a positive recommendation or a positive
assessment of the incoming message). The trust decrease curve illustrates how trust decreases in
effect of negative events (the events that lead to trust decrease, like negative assessment of an
incoming message or receiving a negative recommendation). On the X-axis we have events
affecting trust (positive or negative) and on Y-axis we have trust levels. For instance, from Figure
7 we can see that 7 subsequent negative events are sufficient to drop from full trust (trust level =
1) down to the cut-off level (trust level = 0.2). And we need 25 subsequent positive events to
increase trust from the level = 0.5 to the level 0.9.
The increase of trust in effect of a positive event is bigger if the current level of trust is lower, and
then it gets ‘saturated’ (the subsequent positive events have less impact on trust). In contrary, the
loss of trust (in case of negative events) is much faster and much less dependents on the current
level of trust. In effect, the same numbers of positive and negative events can result in a
significant trust decrease, depending on the current level of trust. And if the number of positive
events grows (comparing to the number of negative events), at some point they can ‘outweigh’
the influence of the negative events resulting in that the isolation of the malicious node is getting
delayed.
Figure 7. Trust level depending on number of events causing trust change
In addition, in the experiment illustrated in Figure 6, the probability ps of spoiling a message by a
malicious node was 90% which means that 10% of the sent messages were non-malicious and
caused the rise of the reputation of the sending node. The result shown in Figure 6 is the
cumulative effect of the influence of the negative events and the positive ones (which number was
growing reverse proportionally to the value of parameter r).
8. CONCLUSIONS
The paper presents WTC2M, a new method of trust management in clustered wireless sensor
networks. The method is based on a distributed trust management model. The main innovation of
WTC2M is that in addition to two trust related mechanisms, namely the local assessment of the
validity of incoming messages and the recommendations received from other nodes it introduces
two additional mechanisms, the deviation history and the validity history, and integrates them all
into a single trust management mechanism able to efficiently counteract the attacks against the
network and against the trust management mechanism. The mechanism takes into account the
histories of local assessments and of the received recommendations (the lengths of these histories

17
are the parameters of the method). It increases the efficiency of WTC2M in situations where the
malicious nodes modify their behaviours to avoid detection. The efficiency of WCT2M in
detecting and isolating the malicious nodes for different attack scenarios has been evaluated in
laboratory and simulation experiments. Four different experiments were conducted, each
involving multiple test cases and multiple simulation runs for each test case.
The considered attacks are representative for a wide range of more specialized attacks. For
instance, the selective forwarding attack (dropping a part or whole of a forwarded message) can
be considered as a combination of black hole attack (dropping a message) and message
modification attack (modifying a message). Node capture is a prerequisite for all attack types
considered in the paper and therefore is a part of each of them.
WCT2M is neutral with respect to network protocols because it is decentralized (cluster heads do
not have any special role in WCT2M based trust management) and is implemented on top of the
stack of protocols. In some cases however, the method can have positive impact on these
protocols, for instance by taking into account the current trust value while voting on a new
cluster-head.
During experiments it was assumed that the nodes use broadcasting to distribute
recommendations. If this is a problem, WCT2M can be easily modified to sending the
recommendations directly instead of broadcasting them. In this research we did not address the
problem of when to switch between broadcasting and direct sending of recommendations (which
can be a topic to be addressed in further research).
There are numerous questions which are in the scope of further research. So far, the experiments
were restricted to the fixed networks and it would be interesting to investigate how WCT2M
performs in mobile networks which may involve clusters reformation. In such situation, malicious
nodes could assume the strategy of changing clusters to follow these where they have the highest
trust values. Another interesting research issue is the gossip problem. As each node is a neighbour
of just few other nodes, it learns about the remaining nodes indirectly, by exchanging trust tables
received as recommendations. It means that most of the nodes know others only from a kind of
‘gossip’. This indirect knowledge is exchanged in both directions, to and from a node, and it
would be interesting to examine how this gossip information may influence the precision and
efficiency of WTC2M. Another issue is to investigate WCT2M behaviour while facing more
intelligent attacks. The present attack models implemented in WCTMS simulator are based on the
probabilistic characterization of network behaviour where the probabilities are static in the sense
that they do not change during simulations (researching towards more intelligent attack models,
including attacks being a combination of several simpler attacks will require further expansion of
WCTMS).
ACKNOWLEDGEMENTS
This work was partially supported by the Statutory Grants 2014-2017 by Polish Ministry of
Science and Higher Education.
REFERENCES
[1] Y. Zhiyng, K. Daeyoung, L. Insun, K. Kiyoung and J. Jongsoo, "A Security Framework with Trust
Management for Sensor Networks," in Proc. Workshop of the 1st International Conference on
Security and Privacy for Emerging Areas in Communication Networks, Athens, IEEE, 2005, pp. 184-
192.
[2] ZigBee Alliance, "Standards: ZigBee Specification," 28 01 2017. [Online]. Available:
http://www.zigbee.org/download/standards-zigbee-specification/. [Accessed 05 11 2017].
[3] IEEE, "Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low
Rate Wireless Personal Area Networks (LR-WPANs)," IEEE Press, New York, 2003.

18
[4] ANGEL, "Description of Final Angel Demonstrator," 6th FR STREP Project: Advanced Networked
embedded platform as a Gateway to Enhance quality of Life (ANGEL), Deliverable 5.2, 2008.
[5] A. Keshavarzian, H. Lee and L. Venkatraman, "Wakeup Scheduling in Wireless Sensor Networks," in
Proceeding of the 7th ACM International Symposium on Mobile ad hoc networking and computing,
Florence, Italy, ACM, 2006, pp. 322-333.
[6] J. Górski and A. Turower, "Assessing the time effectiveness of trust management in fully
synchronised wireless sensor networks," in Design, development and implementation of real-time
systems; Trybus L. & Mastalerz M.W. (eds), Szczecin, Poland, PTI, 2013, pp. 31-42.
[7] N. I. Blatt, "Operational Trust: A New Look at the Human Requirement in Network Centric Warfare,"
in 9th International Command and Control Research and Technology Symposium (ICCRTS):
Coalition Transformation: An Evolution of People, Processes and Technology to Enhance
Interoperability; Leoni Warne (ed), Copenhagen, Danmark, Command and Control Research Program
(US), 2004, pp. 1-17.
[8] J. Górski, A. Turower and A. Wardziński, "Distributed Trust Management Model for Wireless Sensor
Networks," in Monographs of system dependability: Problems of dependability and modelling;
Mazurkiewicz J., Sugier J., Walkowiak T. & Michalska K. (eds), Wrocław, Poland, Oficyna
WYdawnicza Politechniki Wrocławskiej, 2011, pp. 83-94.
[9] T. Zahariadis, H. C. Leligou, P. Trakadas and S. Voliotis, "Trust management in wireless sensor
networks," Transactions on Emerging Telecommunications Technologies, vol. 21, pp. 386-395, 2010.
[10] Y. Yu, K. Li, W. Zhou and P. Li, "Trust mechanisms in wireless sensor networks: Attack analysis and
countermeasures," Journal of Network and Computer Applications, vol. 35, vol. 35, pp. 867-880,
2012.
[11] J. Lopez, R. Roman, I. Agudo and C. Fernandez-Gago, "Trust management systems for wireless
sensor networks: Best practices," Journal of Network and Computer Applications, vol. 35, vol. 33, pp.
1086-1093, 2010.
[12] F. Buccafurri, L. Coppolino, S. D’Antonio, A. Garofalo, G. Lax, A. Nocera and L. Romano, "Trust-
Based Intrusion Tolerant Routing in Wireless Sensor Networks," in Computer Safety, Reliability, and
Security. SAFECOMP 2014, Lecture Notes in Computer Science; Bondavalli A., Di Giandomenico F.
(eds), vol. 1, Florence, Italy, Springer, 2014, pp. 214-229.
[13] R. Abassi and S. G. El Fatmi, "Countering the Collusion Attack in a Trust-based MANET," in
Proceedings of the European, Mediterranean & Middle Eastern Conference on Information Systems
2015; Kostantinos Lambrinoudakis, Vincenzo Morabito & Marinos Themistocleous (eds)., vol. 1,
Athens, Greece, 2015, pp. 575-585.
[14] R. Perlman, "An overview of PKI trust models," IEEE Network, vol. 36, vol. 13, no. 6, pp. 38-43,
1999.
[15] H. Chen, H. Wu, X. Zhou and C. Gao, "Agent-based Trust Model in Wireless Sensor Networks," in
Proc. International Conference on Multimedia and Ubiquitous Engineering, Busan, Korea, IEEE,
2008, pp. 150-154.
[16] H. Chen, H. Wu, X. Zhou and C. Gao, "Reputation-based Trust in Wireless Sensor Networks," in
Proc. International Conference on Multimedia and Ubiquitous Engineering, Seoul, Korea, IEEE,
2007, pp. 603-607.
[17] A. Boukerche and X. Li, "An Agent-based Trust and Reputation Management Scheme for Wireless
Sensor Networks," in Proc. IEEE Global Telecommunications Conference, St. Louis, USA, IEEE,
2005, pp. 1857-1861.
[18] J. Konorski and R. Orlikowski, "DST-Based Detection of Non-cooperative Forwarding Behavior of
MANET and WSN Nodes," in ireless and Mobile Networking. IFIP Advances in Information and
Communication Technology, vol 308, Berlin, Springer, 2009, pp. 185-196.
[19] N. Labraoui, "A reliable trust management scheme in wireless sensor networks," in 12th International
Symposium on Programming and Systems, Algiers, Algeria, IEEE, 2015, pp. 1-6.
[20] W. Fang, W. Zhang, Y. Yang, Y. Liu and W. Chen, "A resilient trust management scheme for
defending against reputation time-varying attacks based on BETA distribution," Science China
Information Sciences, vol. 60, no. 4, pp. 1-11, 2017.

19
[21] C. Karlof and D. Wagner, "Secure Routing in Wireless Sensor Networks: Attacks and
Countermeasures," in First IEEE International Workshop on Sensor Network Protocols and
Applications (SNPA 03), Anchorage, Alaska, IEEE, 2003, pp. 113-127.
[22] Y. Sun, Z. Han and K. Liu, "Defense of Trust Management Vulnerabilities in Distributed Networks,"
Communications Magazine, vol. 46, no. 46, pp. 112-119, 31 March 2008.
[23] B. Mamalis, D. Gavalas, C. Konstantopoulos and G. Pantziou, "Clustering in Wireless Sensor
Networks," in RFID and Sensor Networks: Architectures, Protocols, Security, and Integrations, L. Y.
Y. Zhang (ed.), CRC Press, 2009, pp. 324-353.
[24] Texas Instruments, "CC2520 Development Kit," 10 07 2017. [Online]. Available:
http://www.ti.com/tool/cc2520dk. [Accessed 05 11 2017].
[25] NetBeans, "NetBeans IDE Features," 10 07 2017. [Online]. Available:
https://netbeans.org/features/index.html. [Accessed 05 11 2017].
[26] Java, "Java," 10 07 2017. [Online]. Available: http://www.java.com. [Accessed 05 11 2017].
[27] JGraph Ltd, "JGraph," 10 07 2017. [Online]. Available: http://www.jgraph.com/. [Accessed 05 11
2017].
[28] R. van Solingen and E. Berghout, The Goal/Question/Metric method: A practical guide for quality
improvement of software development, McGraw-Hill Publishing Company, 1999.
[29] A. Turower, "Trust Management Method for Wireless Sensor Networks (Doctoral dissertation),"
2017. [Online]. Available: https://mostwiedzy.pl/publication/download/1/trust-management-method-
for-wireless-sensor-networks_16168.pdf. [Accessed 5 11 2017].
[30] E. Bulut, Z. Wang and B. K. Szymanski, "The Effect of Neighbor Graph Connectivity on Coverage
Redundancy in Wireless Sensor Networks," Communications (ICC), 2010 IEEE International
Conference on, pp. 1 - 5, 2010.
[31] F. Delicato, F. Protti, J. F. De Rezende, L. Rust and L. Pirmez, "Application-driven node management
in multihop wireless sensor networks," Lecture Notes in Computer Science, pp. 569-576, 2005.
[32] A. Jhumka and L. Mottola, "On Consistent Neighborhood Views in Wireless Sensor Networks,"
Reliable Distributed Systems, 2009. SRDS '09. 28th IEEE International Symposium on, pp. 199 -
208, 2009.
[33] W. Ye, J. Heidemann and D. Estrin, "An Energy-Efficient MAC Protocol for Wireless Sensor
Networks," in Proc. INFOCOM 2002. Twenty-First Annual Joint Conference of the IEEE Computer
and Communications Societies, New York, USA, IEEE, 2002, pp. 1567-1576. [2] Gizem, Aksahya &
Ayese, Ozcan (2009) Coomunications & Networks, Network Books, ABC Publishers.
AUTHORS
Janusz Górski is a professor at the Department of Software Engineering, Gdansk
University of Technology in Poland where he is leading the Information Assurance Group
(IAG) http://iag.pg.gda.pl/iag/.
Alan Turower is presently a cloud software developer and a design lead at Intel
Technology Poland company.

A METHOD OF TRUST MANAGEMENT IN WIRELESS SENSOR NETWORKS

Recommended

Recommended

More Related Content

What's hot

What's hot (19)

Similar to A METHOD OF TRUST MANAGEMENT IN WIRELESS SENSOR NETWORKS

Similar to A METHOD OF TRUST MANAGEMENT IN WIRELESS SENSOR NETWORKS (20)

Recently uploaded

Recently uploaded (20)

A METHOD OF TRUST MANAGEMENT IN WIRELESS SENSOR NETWORKS